Published inTowards Deep LearningGoodbye DINO Tricks, Goodbye Stop-Grad: LeJEPA Is About to Rewrite SSL For Computer VisionLeJEPA fixes self-supervised learning with a stable, heuristic-free method that improves in-domain pretraining and model selection.4d ago4d ago
Published inTowards Deep LearningA $4.5 Million LLM Just Beat GPT-5 and Claude: Meet Kimi K2 ThinkingKimi K2 Thinking: The $4.5M Open-Source AI, that’s Challenging GPT-5 and Claude on Reasoning, Speed, and Cost.Nov 10A response icon1Nov 10A response icon1
Published inTowards Deep LearningKimi Linear Just Solved the Million-Token ProblemLet me paint you a picture: You’re trying to process a million tokens. Think an entire codebase, a full novel, or months of conversation…Nov 4A response icon2Nov 4A response icon2
Published inTowards Deep LearningEven AI Can Get Brain Rot: What Happens When LLMs Feed on Junk DataLLMs lose reasoning, memory, and safety when fed junk data. This study shows AI can literally get “brain rot.”Oct 27Oct 27
Published inTowards Deep LearningSamsung Just Built a 7M-Parameter Brain That Outsmarts GiantsSamsung’s 7M-parameter Tiny Recursive Model beats billion-param LLMs on reasoning puzzles like Sudoku, Maze & ARC-AGI.Oct 22A response icon4Oct 22A response icon4
Published inTowards Deep LearningHuggingChat Omni just hijacked 115 models to beat ChatGPTHuggingChat Omni v2: 115 open source models in one chat that auto picks the best LLM per prompt; can it beat ChatGPT now!!Oct 17Oct 17
Published inTowards Deep LearningClaude vs Codex: Here’s What Actually Worked.Hands-on review: Claude Code to architect, OpenAI Codex to implement. Practical tips, pitfalls, and the tool that ships.Oct 17A response icon1Oct 17A response icon1
Published inTowards Deep LearningEmbedding Gemma: Messy sentences to clean topic groups (Tutorial)Cluster text fast with Embedding Gemma + auto‑k, PCA shrink, and robust backends (K‑Means/HDBSCAN/Agglo). Copy‑paste tutorial.Oct 15Oct 15
Published inTowards Deep LearningWhy Everyone Will Want DGX Spark on Their Desk — Yes, EveryoneI just saw this picture today and was amazed, I’ve been waiting for this moment for a long time. (No, it’s not Elon.) It’s that tiny…Oct 14A response icon12Oct 14A response icon12
Published inTowards Deep LearningI Just Cloned My Voice in 3 Seconds: On My Laptop.NeuTTS Air: world’s first realistic on-device voice AI: 3-second cloning, GGUF, runs on CPUs, and sounds shockingly human.Oct 14A response icon2Oct 14A response icon2
